Volcengine TLS LogCollector

设计、校验并验收 TLS LogCollector 采集配置

Adapted from tls-logcollector in volcengine-tls/ve-tls-cli (Apache-2.0). Command examples keep the upstream native volclog ... form; run them as ve volclog ... (requires ve >= 1.1.11) with all later arguments unchanged.

Scope

Use this skill for LogCollector collection tasks. Use volcengine-tls as the generic CLI operating model and this skill as the LogCollector-specific workflow.

First classify the request:

• TLS resource or collector-rule configuration
• Linux host installation or lifecycle management
• Kubernetes DaemonSet installation
• Kubernetes Controller/CRD installation or rule management
• heartbeat, binding, parsing, or ingestion troubleshooting

If the user only asks for a plan or commands, do not change remote resources. If the user asks to execute, show the resolved target and planned writes, validate them, then perform only the authorized changes.

Required inputs

Resolve only what the selected path needs:

• runtime selector: explicit --profile <name>, --secrets-file <path>, or intentionally selected environment credentials
• region and TLS endpoint; never derive one from the other
• public or private network path
• target Project, Topic, host group, and collector rule, by ID or exact name
• input source: host file, container stdout/stderr, or container file
• for Kubernetes: cluster context, namespace, workload type/name, and container selector
• installation method and version when deployment is in scope

Do not request or echo credential values in chat. Do not put AK/SK in skill files, repositories, generated manifests, or shell history.

Contract-first workflow

1. Run volclog --profile <profile> doctor; add --online only when a live connectivity check is needed.
2. Discover exact actions with volclog tool list <group>.
3. Read every selected contract with volclog tool describe <group.action> --view full.
4. Read references/config-validation.md and validate representative samples before persisting a non-trivial parser, path extraction, timestamp, index delimiter, or processor configuration.
5. Read references/tls-resources.md when Project, Topic, host group, collector rule, index, or binding changes are needed.
6. Run write operations with --dry-run first. Dry-run validates local request shape; it does not prove that the service will accept the request or that parsing semantics are correct.
7. Execute the authorized write once. After an ambiguous write result, reconcile with a read action instead of retrying blindly.
8. Follow the environment-specific deployment reference when installation is in scope:
   • references/linux-host.md
   • references/kubernetes-daemonset.md
   • references/kubernetes-controller.md
9. Read references/verification.md and verify process or workload health, machine heartbeat, rule binding, and Topic ingestion.

Collection design rules

• Treat current tool describe output as authoritative for action IDs, fields, enum values, and required properties. Examples in references are starting points, not frozen API contracts.
• Use host-file input only for host paths, container-stdout input only for stdout/stderr, and container-file input only for files inside containers. Do not mix incompatible Paths and ContainerRule shapes.
• Select the narrowest stable path and Kubernetes filters. Never create an empty container filter that can collect the entire cluster by accident.
• Use a Label host group unless the user or an existing deployment explicitly requires IP matching. The installed label and the host group's HostIdentifier must agree. Never configure both label and IP identity on the same agent.
• Reuse existing resources only after exact-name or exact-ID reconciliation. If multiple candidates match, stop and ask the user to select one.
• Do not invent production values for shard count, retention, index settings, resource names, paths, regexes, time zones, or parsing keys.
• Do not automatically collect LogCollector's own logs. A self-log POC is allowed only when the user explicitly requests a temporary end-to-end test and accepts the feedback-loop risk; define teardown and observe two short windows.
• Installation success is not collection success. A healthy process or Pod, machine heartbeat, rule binding, and real Topic ingestion are separate acceptance gates.

Stop conditions

Stop and report the exact stage when:

• runtime selector, region, endpoint, target account, cluster, or resource identity is ambiguous
• a write returns an ambiguous transport result and read-after-write cannot reconcile it
• generated or supplied parsing configuration fails representative sample validation
• installation would overwrite or delete an existing service, DaemonSet, release, CRD, ClusterRole, binding, or collector rule outside the confirmed target
• a POC self-log rule shows accelerating ingestion or resource usage

Output contract

Report:

• selected runtime, target environment, and resource IDs or exact names
• collection source and parser strategy
• validation actions run and their results
• planned and completed writes, separating dry-run from live execution
• deployment commands and where each command runs
• verification evidence for process or Pod health, heartbeat, binding, and Topic ingestion
• cleanup steps for any temporary POC resources